Semiconductor devices (e.g., integrated circuits, displays, and the like) are becoming increasingly dense and highly integrated. With this trend, certain processes such as processes directed to formation of multiple wiring layers, formation of insulating films between wiring layers, formation of inter-layer connecting plugs and the like, and formation of electrodes are becoming increasingly more critical. For example, controlling the thickness and depth profiles of inter-layer insulating films or metal layers is now very important to ensure attainment of target integration levels in devices with high reliability. Achieving such control requires that layer thicknesses in steps such as layer-formation steps and etching steps be monitored.
Responsive to the need to achieve ever-increasing device density along with ever-decreasing feature sizes, the microlithography industry has developed microlithography apparatus that utilize extremely short wavelengths of light, such as deep-UV light, but at large numerical apertures. Projection optics used in such apparatus have extremely short focal ranges. The focal ranges are now so short that the uneven surface that results from stacking multiple layers atop one another during manufacture of an integrated circuit is no longer in sharp focus from peaks to valleys of the surface. Consequently, it has become increasingly important to planarize the surface of the wafer (at least within exposure areas) accurately between certain layer-formation steps. It is also important to perform a planarization step after embedding an inlay of a metal electrode layer to form inter-layer connecting plugs and the like.
Planarization typically involves the removal of material from the surface of the wafer. Whereas several candidate techniques have now been developed for performing planarization, as summarized below, a key problem has been how to accurately detect, during planarization, when to stop planarizing so as to ensure the desired amount of material has been removed without removing excessive material.
Among the several conventional planarization methods, a polishing process termed "chemical mechanical polishing" or "chemical mechanical planarization" (abbreviated "CMP") has received considerable favor. This is because, inter alia, CMP is effective for planarizing wafers having a large surface area and is effective for planarizing microscopic bumps and other surficial irregularities from wafer surfaces. CMP achieves such results from a combination of mechanical abrasion (using an abrasive in liquid suspension) and chemical action (using a mild surface-eroding chemical in the liquid suspension). More specifically, during CMP, the wafer surface is urged against a polishing pad, saturated with a polishing slurry, as the wafer and polishing pad undergo motion relative to each other. The polishing slurry is a suspension of polishing granules (silica, alumina, cerium oxide, or the like, depending on the material on the surface of the wafer) in an acidic or basic (depending on the material on the wafer surface) carrier liquid. With CMP, the entire surface of the wafer can be polished uniformly by making sure that the applied polishing pressure, amount of slurry used, and velocity of relative motions are uniform over the entire wafer surface.
Unfortunately, achieving consistent results with CMP is much more difficult than with other semiconductor processing steps such as layer-forming and etching. Therefore, there is a great need for improved methods for monitoring the thickness of the layer(s) on a wafer being polished by CMP, especially such methods that can provide quick, accurate, and efficient feedback to the CMP apparatus.
According to one conventional approach for monitoring polishing, changes in friction between the wafer and the polishing pad are monitored as corresponding changes in the torque being applied by a motor used to effect rotation of the wafer or polishing pad. For example, a change in torque is encountered when polishing has progressed to an underlying layer made of a material having a substantially different coefficient of friction than the layer being polished away. Unfortunately, monitoring polishing by monitoring torque is notoriously inaccurate and unreliable.
Optical methods for monitoring polishing offer prospects for high accuracy. According to one conventional method, a small wafer "blank" region (i.e., a location on a wafer where the surficial layer is essentially planar and desirably not patterned) is subjected to the same CMP process as the remainder of the wafer and used as a measurement sample. Measuring the thickness of the surficial layer as polishing progresses is performed by monitoring changes in the blank region.
Unfortunately, the "blank"-measurement method has several serious disadvantages. First, the method requires considerable time to execute and to provide feedback to the actual polishing process. Wafers are normally imprinted with as many devices as possible placed side by side. A non-imprinted portion, for use as a blank, of the wafer must be located among the devices on the wafer. The area of such a blank is normally very small and, on some wafers, has an unspecified location. Because the size of the blank region is normally very small, the range of measurements that can be performed at the location is also very small. Available apparatus for accurately measuring layer thickness within such a small area are simply not available. It is also very difficult to perform measurements in such a small area at sufficiently high speed because the required mechanism for picking up, recognizing, and processing the image of the blank region is very complex.
Also, positioning the wafer for measurement at the blank location poses many problems. Because the blank location is normally very small (and sometimes not even specified), accurate alignment mechanisms are required to ensure that the measurements are consistently performed at the blank location.
Another substantial contributor to the excessive amount of time required to perform the "blank" technique is the need to interrupt polishing, clean the wafer, and transport the wafer to a remote but stable location for measurement.
In another conventional optical method for monitoring polishing, the thickness of the layer being polished is monitored by optical interference. In such a technique, an optical path is provided through the polishing pad to the wafer surface being polished, and a laser light beam is directed through the polishing pad to the wafer surface during polishing. Alternatively, light (e.g., infrared light) is transmitted through the wafer holder and through the wafer from the rear surface of the wafer to the surface being polished. Temporal changes in the intensity of light reflected from the surface being polished are monitored as polishing progresses, and a polishing endpoint is believed to be reached when the intensity of the reflected laser light ceases to change with further polishing. Use of such a method for measuring the thickness of a layer at a "blank" location is normally effective in achieving a satisfactory accuracy. However, requisite accuracy is not obtained whenever the method is used to detect a polishing endpoint for a patterned layer (which must be performed when attempting to monitor polishing in real time as polishing progresses). This problem is even more pronounced when the wafer surface is patterned with logic circuits or a combination of logic and memory circuits.
Moreover, because the wafer undergoes motion during polishing, signals from a specific (blank) location on the wafer are not normally obtainable, either directly or extracted from signals from other portions of the wafer surface. This method is also susceptible to obfuscating effects of signal noise, making determinations of polishing endpoint even more difficult to obtain. Consequently, polishing endpoint detection by this technique as conventionally applied cannot be reliably realized, especially in real time.
According to yet another conventional optical approach, the layer thickness is monitored by illuminating onto the wafer surface a light beam having a relatively large spot diameter. Optical measurements of light reflecting from the wafer surface are made and averaged. Unfortunately, if the wafer is imprinted with device patterns, the waveform of reflected light is usually extremely complex since the signal depends not only upon the layer thickness but also on the device patterns. Therefore, the thickness of a layer being polished cannot be measured reliably or accurately using this method.
Other methods that have received some attention include detecting changes in vibration or sounds, or detecting changes in slurry composition that purportedly occur whenever a polishing endpoint is being approached. None of these techniques has proved to be effective or reliable.
As can be ascertained from the foregoing summary, none of the conventional techniques for monitoring the thickness of a layer during polishing is satisfactory for current requirements. Each technique as conventionally applied suffers from one or more deficiencies such as, inter alia, excessive time required for performing the measurements and providing feedback to the polishing process, inaccuracy of determinations of layer thickness, lack of repeatability, and/or inapplicability to certain types of layers or to certain steps during manufacture of a semiconductor device.